research-agent

Research Agent Domain

Research the domain expertise needed to write a high-quality agent definition. Produces a research file that the generate-agent skill consumes.

Input

$ARGUMENTS is the agent name (e.g., data-audits, eval-governance, synth-debate).

Process

Step 1: Read the plan

The calling skill passes the agent's config (purpose, sources/dimensions, research needed) via the prompt. No plan file to read.

Find the section for $ARGUMENTS. Extract:

• Purpose — what this agent does
• Sources — specific data sources or APIs (for data agents)
• Dimensions — scoring dimensions (for eval agents)
• Research needed — what domain knowledge to gather

Step 2: Research

Based on the agent type (determined by prefix):

For data-* agents:

• WebSearch for the specific data sources mentioned in the plan
• WebFetch key pages to understand data format and availability
• Look for: API documentation, data schemas, access methods, rate limits
• Look for: alternative sources, cross-referencing approaches

For eval-* agents:

• WebSearch for evaluation methodologies in this domain
• WebFetch academic or practitioner frameworks for scoring
• Look for: established scoring rubrics, industry benchmarks, common pitfalls
• Look for: what distinguishes excellent from adequate in each dimension

For synth-* agents:

• WebSearch for synthesis and decision-making frameworks
• Look for: how expert panels aggregate opinions, handling disagreement
• Look for: report formats that decision-makers actually use

Step 3: Write research file

Write findings to research/$ARGUMENTS.md:

# Research: $ARGUMENTS

**Researched:** YYYY-MM-DD
**Purpose:** [from plan]

## Domain Context

[2-3 paragraphs of domain background relevant to this agent's role]

## Data Sources Found

[For data agents: specific URLs, APIs, access methods]
[For eval agents: frameworks, rubrics, benchmarks discovered]
[For synth agents: synthesis methodologies, report formats]

### Source 1: [name]
- **URL:** [url]
- **What it provides:** [description]
- **Access method:** [API/scrape/manual]
- **Reliability:** [high/medium/low]

### Source 2: [name]
...

## Methodology Notes

[For eval agents: how to score each dimension based on research]
[For data agents: how to normalize data across sources]
[For synth agents: how to handle disagreement, weighting]

## Key Findings

- [Finding 1 — something that should influence the agent definition]
- [Finding 2]
- [Finding 3]

## Gaps

[What couldn't be found — this is valuable for setting agent expectations]